Weighted Feature Selection Techniques for Detecting Impersonation Attack in Wi-Fi Networks

As Internet-of-Things (IoT) devices enable pervasive computing in our daily lives, more and more devices are connected to Wi-Fi networks. The public access to Wi-Fi network leads to exploitable vulnerabilities that can be inverted as attacks. Impersonation attack is an active malicious action where unauthorized users masquerade themself as authorized to gain privileges. Detecting impersonation attacks remains a challenging task due to its similar properties with benign packets. Moreover, the pervasiveness of IoT devices connected to a Wi-Fi network generates a complex, largescale, and high-dimensional data, which leads to difficulties in real-time detection and mitigation. Selecting the best features is one of the challenging issues to improve the performance of the classifier. In this study, we examine the feature weighting methods of existing machine learners and how they could be used for the accurate selection for impersonation attack features. We test and validate the utility and usefulness of the selected features using a standard neural network. This study finally demonstrates that the proposed weight-based machine learning model can outperform other filterbased feature selection models. We evaluated the proposed model on a well-referenced Wi-Fi networks benchmark dataset, namely, AWID. The experimental results not only demonstrate the effectiveness of the proposed model achieving an accuracy of 99.86% but also prove that combining a weightbased feature selection method with a light machine-learning classifier leads to a significantly better performance compared to the best result reported in the literature.